Physiological sensing of personal thermal comfort with wearable devices in fan-assisted cooling environments in the tropics

Thermal comfort prediction with physiological parameters has been getting increasing attention due to the advances in wearable sensing technology. Previous studies in chamber and air-conditioning environments indicate that physiological parameter-based group and personal comfort models can predict thermal comfort accurately. To demonstrate whether physiological signals are reliable indicators for thermal comfort prediction in fan-assisted cooling environments, a series of experiments were conducted to collect participants’ physiological and thermal responses in a mixed-mode fan-assisted cooling environment in tropical Singapore. Group models and personal comfort models with different machine learning algorithms were then developed. The results show that the accuracy ranges of group thermal comfort models based on all measured physiological features for thermal sensation vote, thermal preference, and air velocity preference predictions are (62.4%, 73.3%), (74.5%, 82.2%), and (67.8%, 77.7%), respectively. For personal comfort models (PCMs), PCMs with all physiological features as inputs have a median accuracy/Area Under the Curve (AUC) of 82.0%/0.92, 84.5%/0.92, and 80.7%/0.91 for TSV, TP, and VP prediction, respectively. Additionally, personal comfort models based on four groups of input features were developed and compared to explore the feasibility of using fewer physiological parameters to predict thermal comfort. Finally, this study demonstrates that only using two skin temperatures from wearable body parts can predict thermal comfort accurately in fan-assisted cooling thermal environments.